search

thangvubk / SoftGroup

[CVPR 2022 Oral] SoftGroup for Instance Segmentation on 3D Point Clouds
MIT License
341 stars 80 forks source link

S3DIS pretrained model inference #101

Open zemovi opened 2 years ago

zemovi commented 2 years ago

Hey, thanks for the method. A simple inference on the provided S3DIS pretrained model on Area_1* gives AP of 0.003? python tools/test.py configs/softgroup_s3dis_fold5.yaml work_dirs/softgroup_s3dis_spconv2.pth --out results issue_github_softgroup

thangvubk commented 1 year ago

I think 0.003 is unexpected. Did you make any modification? Could you test the model from the clean lastest commit?

Naruto-xsl commented 1 year ago

The name of S3DIS dataset is Standard3dDataset v1.2 Aligned_ Version folder?

ImaneTopo commented 1 month ago

Hey, thanks for the method. A simple inference on the provided S3DIS pretrained model on Area_1* gives AP of 0.003? python tools/test.py configs/softgroup_s3dis_fold5.yaml work_dirs/softgroup_s3dis_spconv2.pth --out results issue_github_softgroup

Hi, I want to ask you if you wonder, how did you achieve training the model on S3DIS, because in my case, I want to train the model from the scratch on the dataset s3dis, the traning was done, but in the validation phase, I face an erreor because the dict res were imcomplete due to the absence of pred_instances and gt_instances.

Capture11